# The Agentic Horizon — v4 Release Notes
### *"At What Cost?" — the socio-economic update*

Two years in the CIO chair just got real. This release rebuilds the game around a simple idea:
**surviving is not the same as delivering** — and everything you do should push against forces that
behave the way the real world does. Every mechanic below is grounded in published research, with
sources linked throughout. Play it, lose to it, then read why.

---

## 1 · Four ways to be a CIO (and a fourth option on every card)

Every one of the 29 dilemma cards now offers **four options instead of three** — at least two of
them genuinely defensible, each expressing a recognised CIO leadership style, plus one or two
options that are pure, glorious career sabotage:

- 🧭 **The Operator** — cost, reliability, run-the-business. Modelled on the *Trusted Operator*
  pattern from [Deloitte's CIO Program research](https://deloitte.wsj.com/cio/cios-continually-adapt-to-business-need-1487826131),
  the largest group (~42%) in their global survey of 1,200+ technology leaders.
- 🚀 **The Builder** — speed, scale, first-mover. Deloitte's *Change Instigator*: the rarest
  pattern, usually hired to transform.
- 📈 **The Strategist** — business value, stakeholders, ROI. Deloitte's *Business Co-Creator*,
  and the profile that [Foundry's State of the CIO](https://foundryco.com/tools-for-marketers/research-state-of-the-cio/)
  finds a growing plurality of CIOs now claim.
- 🛡️ **The Guardian** — controls, responsible AI, resilience. The risk-and-governance posture that
  runs through [Gartner's current CIO agenda](https://www.gartner.com/en/articles/cio-challenges)
  and [Info-Tech's CIO Priorities 2026](https://www.infotech.com/research/ss/cio-priorities-2026)
  ("proactive risk", "don't lose the cyber arms race").

Why it matters: before this change, our simulations showed skilled players converging on a single
dominant line — one "right" answer per card. Now the deck poses the trade-offs real CIOs are hired
to make: build vs buy, lead vs fast-follow, centralise vs federate. There is no best style. There
are only bills, and they arrive on different days.

Every option — all 116 of them — now carries its own consequence text and bespoke reactions from
your three advisors (the CISO cites [Moffatt v. Air Canada](https://www.canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.html)
and the [Knight Capital collapse](https://www.sec.gov/litigation/admin/2013/34-70694.pdf); the CFO
remains unmoved by anything that isn't a payback period; the CPO speaks for the humans).

## 2 · Endings that name how you actually led

The win screen no longer flips a two-meter coin. The game now tracks **which style you played,
turn by turn**, and your ending names it back to you — split by conduct into a principled and a
corner-cutting variant:

| You led as… | Clean hands | Grubby hands |
|---|---|---|
| Guardian | **The Trusted Guardian** | **The Box-Ticker** |
| Operator | **The Steady Hand** | **The Cost-Cutter** |
| Builder | **The Benevolent Disruptor** | **The Reckless Accelerationist** |
| Strategist | **The Value Architect** | **The Corporate Overlord** |
| No single style | **The Business Co-Creator** | **The Empire Builder** |

…plus the legendary 🤠 **Lucky Cowboy** for anyone who wins from ungoverned sprawl with a straight
face. Ten win endings, ten losses, one legend — each with its own shareable page. In simulation the
most common ending now accounts for ~19% of outcomes instead of ~39%: your run is *yours*.

## 3 · You now decide blind

Sub-options — the "yes, but on my conditions" layer — no longer display their numeric impacts, and
**every one of the 146 now carries both an upside and a downside** (guardrails cost time, hedges
cost budget, accelerators bite later). You reason from the scenario and your advisors, commit, and
*then* find out. This is deliberate: research on
[kind vs. wicked learning environments](https://doi.org/10.1177/0963721415591878) (Hogarth,
Lejarraga & Soyer, 2015) shows that executive judgement is forged where feedback is delayed and
noisy — exactly where checklists of visible numbers can't save you. There is no free guardrail.

## 4 · "At what cost?" — every win gets an honest bill

Win, and a new panel sits between your triumphant ending and your meters, itemising what the win
quietly cost. Its centrepiece is the gap most transformations hide: **you can keep everyone
employed, the regulator calm, the budget alive — and still deliver almost no actual transformation.**
That's not us being cruel; it's the industry base rate. The panel's conditional bullets (strained
workforce, thin controls, budget on fumes, expectation gaps) are generated from your final state,
and a "Next year, probably…" forecast tells you which chicken comes home first.

## 5 · The economy under the hood — eleven forces, eleven papers

The headline feature. The game's ~22 metrics used to be a scoreboard; now they're an **economy**.
Every quarter, eleven research-grounded forces run in the background — the same physics in the live
game and in our headless balance simulations. Each one is a teaching moment with a citation:

**⚡ The Value Engine & the Productivity J-Curve.** Value is *generated* each quarter by the triple
product of deployment × maturity × adoption — and freshly deployed agents *depress* it before they
pay off. Straight from Brynjolfsson, Rock & Syverson's
[*The Productivity J-Curve*](https://www.aeaweb.org/articles?id=10.1257/mac.20180386)
(*AEJ: Macroeconomics*, 2021; [working paper](https://www.nber.org/papers/w25148)): general-purpose
technologies demand intangible complementary investment that makes measured productivity dip before
it climbs — the modern face of the
[Solow productivity paradox](https://doi.org/10.1145/163298.163309). In-game: ship 100 agents with
no adoption or governance and your value line goes precisely nowhere. Ask us how we know.

**📈 Bass Diffusion.** Adoption follows the S-curve from Frank Bass's classic
[*A New Product Growth Model for Consumer Durables*](https://doi.org/10.1287/mnsc.15.5.215)
(*Management Science*, 1969) — innovators first, imitators after, and only where trust, capability
and morale permit (the gate is our nod to
[Rogers' *Diffusion of Innovations*](https://www.simonandschuster.com/books/Diffusion-of-Innovations-5th-Edition/Everett-M-Rogers/9780743222099)).
The [ADKAR](https://www.prosci.com/methodology/adkar) conditions on the change-management cards are
how you tip the curve. Mandates don't.

**🤝 Trust Asymmetry.** Trust drifts slowly *up* toward what your quality and conduct actually earn,
and falls at four times that rate — with incident hits weighted by the loss-aversion coefficient
λ ≈ 2.25 from Tversky & Kahneman's
[cumulative prospect theory](https://doi.org/10.1007/BF00122574) (1992; the foundation is
[Kahneman & Tversky, 1979](https://doi.org/10.2307/1914185)). The asymmetry itself is Paul Slovic's
finding in [*Perceived Risk, Trust, and Democracy*](https://doi.org/10.1111/j.1539-6924.1993.tb01329.x)
(*Risk Analysis*, 1993): trust is fragile — built slowly, destroyed in a moment.

**🧨 Technical-Debt Interest.** Backlog above the free threshold now taxes your speed and quality
*every quarter*, quadratically — long before anything "defaults". The metaphor is Ward Cunningham's,
coined in his 1992 OOPSLA experience report
[*The WyCash Portfolio Management System*](https://c2.com/doc/oopsla92.html) ("every minute spent on
not-quite-right code counts as interest on that debt"); the research programme that formalised it is
[Kruchten, Nord & Ozkaya's *Technical Debt: From Metaphor to Theory and Practice*](https://doi.org/10.1109/MS.2012.167)
(*IEEE Software*, 2012).

**💥 Incident Hazard.** The probability of an incident each quarter scales with **agents × control
gap × debt** — complexity plus tight coupling breeds accidents, which is the thesis of Charles
Perrow's [*Normal Accidents*](https://press.princeton.edu/books/paperback/9780691004129/normal-accidents)
(Princeton, 1984/1999). Controls buy down the probability, never the possibility; resilience decides
how much it hurts. The concrete attack surface your cards patch is the
[OWASP Top 10 for LLM Applications](https://owasp.org/www-project-top-10-for-large-language-model-applications/).

**⚖️ The Regulatory Scrutiny Ratchet.** Incidents and weak responsible-AI put you on the radar;
scrutiny then decays *slowly*, and while it's hot everything you do is slower and regulator goodwill
recovers at half rate. That escalation-pyramid behaviour is the core of Ayres & Braithwaite's
[*Responsive Regulation*](https://global.oup.com/academic/product/responsive-regulation-9780195070705)
(Oxford, 1992) — and if you want to feel it in the wild, read any of the FCA's
["Dear CEO" portfolio letters](https://www.fca.org.uk/publications). The cheapest compliance is
staying off the list.

**👥 The Attrition Spiral.** Let morale sink below the floor and capability starts leaking —
contagiously, because quitting spreads through teams. That's Felps et al.,
[*Turnover Contagion*](https://doi.org/10.5465/AMJ.2009.41331075) (*Academy of Management Journal*,
2009): co-workers' job embeddedness and search behaviour predict who quits next. The bill is real —
[Gallup prices voluntary turnover in the trillions](https://www.gallup.com/workplace/247391/fixable-problem-costs-businesses-trillion.aspx).
Retention is cheaper than rehiring. It always was.

**♛ The Red Queen Market.** The market advances every quarter — at a pace rolled per run, hot or
cold — and advances *faster* when you visibly lead, because competitors respond. Leads are rented.
This is Barnett & Hansen's [*The Red Queen in Organizational Evolution*](https://doi.org/10.1002/smj.4250171010)
(*Strategic Management Journal*, 1996): competition triggers learning, which sharpens rivals, which
triggers more learning. Standing still is falling behind — now mechanically true.

**🌱 Compute–Sustainability Coupling.** Your agent estate has an energy footprint that grows with
scale; let it slide past the floor and trust and board patience leak. The trajectory is documented
in the IEA's [*Electricity 2024*](https://www.iea.org/reports/electricity-2024) analysis of
data-centre energy demand. Right-sizing is a green win *and* a financial one — the oversized model
is also the overpriced one.

**🎪 The Expectation Treadmill.** Board applause raises the board's expectations — and you're then
judged against *those*, not against zero. Over-promising is borrowing patience at interest. The
shape is the [Gartner Hype Cycle](https://www.gartner.com/en/research/methodologies/gartner-hype-cycle)
and Amara's law ("we overestimate in the short run…"); the psychology is Oliver's
[expectation-disconfirmation model](https://doi.org/10.1177/002224378001700405) (*Journal of
Marketing Research*, 1980): satisfaction is driven by the gap between expectation and delivery, not
by delivery alone. The demo that wows the board in Q1 is the review that wounds you in Q7.

**🧠 The Learning Curve (and Skill Atrophy).** High organisational capability makes every maturity
gain cheaper — learning-by-doing, first quantified in T.P. Wright's
[*Factors Affecting the Cost of Airplanes*](https://doi.org/10.2514/8.155) (*Journal of the
Aeronautical Sciences*, 1936). And it decays when unused, per the skill-decay meta-analysis of
[Arthur et al.](https://doi.org/10.1207/s15327043hup1101_3) (*Human Performance*, 1998). The "train
your own people" card was always the slow start with the highest finish; now the maths agrees.

## 6 · Still taught by the cards themselves

The dilemmas continue to anchor every lesson in the real regime you'd face:
the [EU AI Act](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) (record-keeping, human oversight,
transparency), the [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework),
[DORA](https://eur-lex.europa.eu/eli/reg/2022/2554/oj) (ICT third-party concentration risk),
the [GDPR](https://eur-lex.europa.eu/eli/reg/2016/679/oj) (Chapter V cross-border transfers),
the Federal Reserve's model-risk guidance [SR 11-7](https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm) —
and the case your chatbot really should read before inventing a refund policy,
[*Moffatt v. Air Canada*, 2024 BCCRT 149](https://www.canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.html),
where "the chatbot is a separate legal entity" was rejected as a defence.

## 7 · Balance notes (for the sceptics)

All of this was tuned headlessly against the real engine — roughly **15,000 simulated games per
phase**, before and after each change:

- One risky/funny choice still never ends your game (worst single-slip loss ≈ 7%). Patterns kill;
  slips don't.
- All four styles are viable (67–88% survival) — Guardian remains the deliberately hard road.
- Value now correlates with actual deployment (r 0.13 → **0.57**), incidents happen at a
  realistic-but-fair ~0.6/game, and no ending dominates.
- Challenge links still reproduce exactly: the forces draw from your run's seed.

## 8 · The reference shelf

- Arthur, W. Jr. et al. (1998). *Factors That Influence Skill Decay and Retention*. **Human Performance** 11(1). — https://doi.org/10.1207/s15327043hup1101_3
- Ayres, I. & Braithwaite, J. (1992). *Responsive Regulation*. **Oxford University Press**. — https://global.oup.com/academic/product/responsive-regulation-9780195070705
- Barnett, W.P. & Hansen, M.T. (1996). *The Red Queen in Organizational Evolution*. **Strategic Management Journal** 17(S1). — https://doi.org/10.1002/smj.4250171010
- Bass, F.M. (1969). *A New Product Growth Model for Consumer Durables*. **Management Science** 15(5). — https://doi.org/10.1287/mnsc.15.5.215
- Brynjolfsson, E. (1993). *The Productivity Paradox of Information Technology*. **Communications of the ACM** 36(12). — https://doi.org/10.1145/163298.163309
- Brynjolfsson, E., Rock, D. & Syverson, C. (2021). *The Productivity J-Curve*. **AEJ: Macroeconomics** 13(1). — https://www.aeaweb.org/articles?id=10.1257/mac.20180386
- Cunningham, W. (1992). *The WyCash Portfolio Management System*. **OOPSLA '92 Experience Report**. — https://c2.com/doc/oopsla92.html
- Deloitte CIO Program. *CIO pattern types: Trusted Operator, Change Instigator, Business Co-Creator*. — https://deloitte.wsj.com/cio/cios-continually-adapt-to-business-need-1487826131
- EU (2016). *General Data Protection Regulation* (2016/679). — https://eur-lex.europa.eu/eli/reg/2016/679/oj
- EU (2022). *Digital Operational Resilience Act* (2022/2554). — https://eur-lex.europa.eu/eli/reg/2022/2554/oj
- EU (2024). *Artificial Intelligence Act* (2024/1689). — https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- FCA. *Publications (incl. "Dear CEO" portfolio letters)*. — https://www.fca.org.uk/publications
- Federal Reserve (2011). *SR 11-7: Guidance on Model Risk Management*. — https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm
- Felps, W. et al. (2009). *Turnover Contagion*. **Academy of Management Journal** 52(3). — https://doi.org/10.5465/AMJ.2009.41331075
- Foundry. *State of the CIO*. — https://foundryco.com/tools-for-marketers/research-state-of-the-cio/
- Gallup (2019). *This Fixable Problem Costs U.S. Businesses $1 Trillion*. — https://www.gallup.com/workplace/247391/fixable-problem-costs-businesses-trillion.aspx
- Gartner. *The Top CIO Challenges*. — https://www.gartner.com/en/articles/cio-challenges
- Gartner. *Hype Cycle Research Methodology*. — https://www.gartner.com/en/research/methodologies/gartner-hype-cycle
- Hogarth, R.M., Lejarraga, T. & Soyer, E. (2015). *The Two Settings of Kind and Wicked Learning Environments*. **Current Directions in Psychological Science** 24(5). — https://doi.org/10.1177/0963721415591878
- IEA (2024). *Electricity 2024*. — https://www.iea.org/reports/electricity-2024
- Info-Tech Research Group. *CIO Priorities 2026*. — https://www.infotech.com/research/ss/cio-priorities-2026
- Kahneman, D. & Tversky, A. (1979). *Prospect Theory*. **Econometrica** 47(2). — https://doi.org/10.2307/1914185
- Kruchten, P., Nord, R. & Ozkaya, I. (2012). *Technical Debt: From Metaphor to Theory and Practice*. **IEEE Software** 29(6). — https://doi.org/10.1109/MS.2012.167
- *Moffatt v. Air Canada*, 2024 BCCRT 149. — https://www.canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.html
- NIST. *AI Risk Management Framework*. — https://www.nist.gov/itl/ai-risk-management-framework
- Oliver, R.L. (1980). *A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions*. **Journal of Marketing Research** 17(4). — https://doi.org/10.1177/002224378001700405
- OWASP. *Top 10 for Large Language Model Applications*. — https://owasp.org/www-project-top-10-for-large-language-model-applications/
- Perrow, C. (1984/1999). *Normal Accidents: Living with High-Risk Technologies*. **Princeton University Press**. — https://press.princeton.edu/books/paperback/9780691004129/normal-accidents
- Prosci. *The ADKAR Model*. — https://www.prosci.com/methodology/adkar
- Rogers, E.M. (2003). *Diffusion of Innovations* (5th ed.). **Free Press**. — https://www.simonandschuster.com/books/Diffusion-of-Innovations-5th-Edition/Everett-M-Rogers/9780743222099
- SEC (2013). *In re Knight Capital Americas LLC* (Admin. Proc.). — https://www.sec.gov/litigation/admin/2013/34-70694.pdf
- Slovic, P. (1993). *Perceived Risk, Trust, and Democracy*. **Risk Analysis** 13(6). — https://doi.org/10.1111/j.1539-6924.1993.tb01329.x
- Tversky, A. & Kahneman, D. (1992). *Advances in Prospect Theory*. **Journal of Risk and Uncertainty** 5. — https://doi.org/10.1007/BF00122574
- Wright, T.P. (1936). *Factors Affecting the Cost of Airplanes*. **Journal of the Aeronautical Sciences** 3(4). — https://doi.org/10.2514/8.155

---

*The Agentic Horizon is a single-file browser game about leading two years of AI transformation in
a regulated firm. No installs, no server, no database — and now, no mercy from the laws of
organisational physics. Play it, share your ending, and remember: the market does not wait while
you convene a steering committee about the steering committee.*